Claude Opus 4.5

Claude Opus 4.5 is Anthropic's flagship release aimed squarely at coding, autonomous agents, and computer-use workloads, arriving in direct response to competitive pressure from OpenAI's GPT-5.1-Codex-Max and Google's Gemini 3. Anthropic positions the model as state-of-the-art on SWE-bench Verified for real-world software engineering, using that benchmark as the headline proof point for its agentic coding claims. The release also emphasizes practical improvements in everyday research and productivity tasks like working with slides and spreadsheets, broadening the model's appeal beyond pure software engineering teams.

A notable shift in Opus 4.5 is the introduction of an adjustable "effort" parameter, defaulting to high but tunable to medium or low for faster responses, giving developers finer control over latency versus thoroughness. The model supports enhanced computer use and ships alongside new developer platform tools designed for longer-running agents. With a 200,000-token context window and 64,000-token output ceiling, it matches Sonnet 4.5's capacity while delivering Opus-level intelligence at a dramatically reduced $5/$the listed price per million token price point, roughly one-third the cost of the previous Opus generation.